Blame view

lib/raid6/mmx.c 3.81 KB
1da177e4c   Linus Torvalds   Linux-2.6.12-rc2
1
2
3
4
5
6
7
  /* -*- linux-c -*- ------------------------------------------------------- *
   *
   *   Copyright 2002 H. Peter Anvin - All Rights Reserved
   *
   *   This program is free software; you can redistribute it and/or modify
   *   it under the terms of the GNU General Public License as published by
   *   the Free Software Foundation, Inc., 53 Temple Place Ste 330,
93ed05e2a   Atsushi SAKAI   md: fix typo in F...
8
   *   Boston MA 02111-1307, USA; either version 2 of the License, or
1da177e4c   Linus Torvalds   Linux-2.6.12-rc2
9
10
11
12
13
   *   (at your option) any later version; incorporated herein by reference.
   *
   * ----------------------------------------------------------------------- */
  
  /*
a8e026c78   NeilBrown   Further tidyup of...
14
   * raid6/mmx.c
1da177e4c   Linus Torvalds   Linux-2.6.12-rc2
15
16
17
   *
   * MMX implementation of RAID-6 syndrome functions
   */
ca5cd877a   Al Viro   x86 merge fallout...
18
  #if defined(__i386__) && !defined(__arch_um__)
1da177e4c   Linus Torvalds   Linux-2.6.12-rc2
19

f701d589a   Dan Williams   md/raid6: move ra...
20
  #include <linux/raid/pq.h>
a8e026c78   NeilBrown   Further tidyup of...
21
  #include "x86.h"
1da177e4c   Linus Torvalds   Linux-2.6.12-rc2
22

a8e026c78   NeilBrown   Further tidyup of...
23
  /* Shared with raid6/sse1.c */
1da177e4c   Linus Torvalds   Linux-2.6.12-rc2
24
25
26
27
28
29
30
31
  const struct raid6_mmx_constants {
  	u64 x1d;
  } raid6_mmx_constants = {
  	0x1d1d1d1d1d1d1d1dULL,
  };
  
  static int raid6_have_mmx(void)
  {
1da177e4c   Linus Torvalds   Linux-2.6.12-rc2
32
33
  	/* Not really "boot_cpu" but "all_cpus" */
  	return boot_cpu_has(X86_FEATURE_MMX);
1da177e4c   Linus Torvalds   Linux-2.6.12-rc2
34
35
36
37
38
39
40
41
42
43
  }
  
  /*
   * Plain MMX implementation
   */
  static void raid6_mmx1_gen_syndrome(int disks, size_t bytes, void **ptrs)
  {
  	u8 **dptr = (u8 **)ptrs;
  	u8 *p, *q;
  	int d, z, z0;
1da177e4c   Linus Torvalds   Linux-2.6.12-rc2
44
45
46
47
  
  	z0 = disks - 3;		/* Highest data disk */
  	p = dptr[z0+1];		/* XOR parity */
  	q = dptr[z0+2];		/* RS syndrome */
a723406c4   H. Peter Anvin   [PATCH] md: RAID6...
48
  	kernel_fpu_begin();
1da177e4c   Linus Torvalds   Linux-2.6.12-rc2
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
  
  	asm volatile("movq %0,%%mm0" : : "m" (raid6_mmx_constants.x1d));
  	asm volatile("pxor %mm5,%mm5");	/* Zero temp */
  
  	for ( d = 0 ; d < bytes ; d += 8 ) {
  		asm volatile("movq %0,%%mm2" : : "m" (dptr[z0][d])); /* P[0] */
  		asm volatile("movq %mm2,%mm4");	/* Q[0] */
  		for ( z = z0-1 ; z >= 0 ; z-- ) {
  			asm volatile("movq %0,%%mm6" : : "m" (dptr[z][d]));
  			asm volatile("pcmpgtb %mm4,%mm5");
  			asm volatile("paddb %mm4,%mm4");
  			asm volatile("pand %mm0,%mm5");
  			asm volatile("pxor %mm5,%mm4");
  			asm volatile("pxor %mm5,%mm5");
  			asm volatile("pxor %mm6,%mm2");
  			asm volatile("pxor %mm6,%mm4");
  		}
  		asm volatile("movq %%mm2,%0" : "=m" (p[d]));
  		asm volatile("pxor %mm2,%mm2");
  		asm volatile("movq %%mm4,%0" : "=m" (q[d]));
  		asm volatile("pxor %mm4,%mm4");
  	}
a723406c4   H. Peter Anvin   [PATCH] md: RAID6...
71
  	kernel_fpu_end();
1da177e4c   Linus Torvalds   Linux-2.6.12-rc2
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
  }
  
  const struct raid6_calls raid6_mmxx1 = {
  	raid6_mmx1_gen_syndrome,
  	raid6_have_mmx,
  	"mmxx1",
  	0
  };
  
  /*
   * Unrolled-by-2 MMX implementation
   */
  static void raid6_mmx2_gen_syndrome(int disks, size_t bytes, void **ptrs)
  {
  	u8 **dptr = (u8 **)ptrs;
  	u8 *p, *q;
  	int d, z, z0;
1da177e4c   Linus Torvalds   Linux-2.6.12-rc2
89
90
91
92
  
  	z0 = disks - 3;		/* Highest data disk */
  	p = dptr[z0+1];		/* XOR parity */
  	q = dptr[z0+2];		/* RS syndrome */
a723406c4   H. Peter Anvin   [PATCH] md: RAID6...
93
  	kernel_fpu_begin();
1da177e4c   Linus Torvalds   Linux-2.6.12-rc2
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
  
  	asm volatile("movq %0,%%mm0" : : "m" (raid6_mmx_constants.x1d));
  	asm volatile("pxor %mm5,%mm5");	/* Zero temp */
  	asm volatile("pxor %mm7,%mm7"); /* Zero temp */
  
  	for ( d = 0 ; d < bytes ; d += 16 ) {
  		asm volatile("movq %0,%%mm2" : : "m" (dptr[z0][d])); /* P[0] */
  		asm volatile("movq %0,%%mm3" : : "m" (dptr[z0][d+8]));
  		asm volatile("movq %mm2,%mm4"); /* Q[0] */
  		asm volatile("movq %mm3,%mm6"); /* Q[1] */
  		for ( z = z0-1 ; z >= 0 ; z-- ) {
  			asm volatile("pcmpgtb %mm4,%mm5");
  			asm volatile("pcmpgtb %mm6,%mm7");
  			asm volatile("paddb %mm4,%mm4");
  			asm volatile("paddb %mm6,%mm6");
  			asm volatile("pand %mm0,%mm5");
  			asm volatile("pand %mm0,%mm7");
  			asm volatile("pxor %mm5,%mm4");
  			asm volatile("pxor %mm7,%mm6");
  			asm volatile("movq %0,%%mm5" : : "m" (dptr[z][d]));
  			asm volatile("movq %0,%%mm7" : : "m" (dptr[z][d+8]));
  			asm volatile("pxor %mm5,%mm2");
  			asm volatile("pxor %mm7,%mm3");
  			asm volatile("pxor %mm5,%mm4");
  			asm volatile("pxor %mm7,%mm6");
  			asm volatile("pxor %mm5,%mm5");
  			asm volatile("pxor %mm7,%mm7");
  		}
  		asm volatile("movq %%mm2,%0" : "=m" (p[d]));
  		asm volatile("movq %%mm3,%0" : "=m" (p[d+8]));
  		asm volatile("movq %%mm4,%0" : "=m" (q[d]));
  		asm volatile("movq %%mm6,%0" : "=m" (q[d+8]));
  	}
a723406c4   H. Peter Anvin   [PATCH] md: RAID6...
127
  	kernel_fpu_end();
1da177e4c   Linus Torvalds   Linux-2.6.12-rc2
128
129
130
131
132
133
134
135
136
137
  }
  
  const struct raid6_calls raid6_mmxx2 = {
  	raid6_mmx2_gen_syndrome,
  	raid6_have_mmx,
  	"mmxx2",
  	0
  };
  
  #endif